Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

AA-LCR

Artificial Analysis Long Context Reasoning (AA-LCR) Dataset includes 100 hard text-based questions that require reasoning across multiple real-world documents, with each document set averaging ~100k input tokens. Questions are designed such that answers cannot be directly retrieved from documents and must instead be reasoned from multiple information sources.
Source:

Model Performance

#1
GPT-5 (Thinking)
76.0%
#2
GPT-5.2 Codex
76.0%
#3
GPT-5
76.0%
#4
Claude Opus 4.5 (Thinking)
74.0%
#5
GPT-5.3 Codex
74.0%
#6
Gemini 3.1 Pro Preview
73.0%
#7
Claude Opus 4.6 (Thinking)
71.0%
#8
Claude Haiku 4.5 (Thinking)
70.0%
#9
OpenAI o3 (Medium Effort)
69.0%
#10
OpenAI o3 (High Effort)
69.0%
#11
MiniMax M2.7
69.0%
#12
GPT-5 Mini (Thinking)
68.0%
#13
Grok 4 (Thinking)
68.0%
#14
Grok 4
68.0%
#15
Qwen 3 (Thinking)
67.0%
#16
GPT-5.1 Codex Max
67.0%
#17
Claude 4.1 Opus (Thinking)
66.0%
#18
MiniMax M2.5
66.0%
#19
Gemini 2.5 Pro (Thinking)
66.0%
#20
Claude Sonnet 4.5 (Thinking)
66.0%
#21
Gemini 2.5 Pro
66.0%
#22
DeepSeek V3.2 (Thinking)
65.0%
#23
Claude 4.0 Sonnet (Thinking)
65.0%
#24
Gemini 3.1 Flash Lite Preview
65.0%
#25
Claude Opus 4.5
65.0%
#26
Gemini 2.5 Flash (Thinking)
62.0%
#27
Claude 3.7 Sonnet (Thinking)
61.0%
#28
GPT-4.1
61.0%
#29
Kimi K2.5
59.0%
#30
OpenAI o1
59.0%
#31
Grok 4.20 (Reasoning)
59.0%
#32
MiniMax M2.1
59.0%
#33
Qwen 3 Max (Thinking)
58.0%
#34
Claude Sonnet 4.6
58.0%
#35
Claude Opus 4.6
58.0%
#36
Grok 3
55.0%
#37
DeepSeek R1
55.0%
#38
OpenAI o4 Mini (Medium Effort)
55.0%
#39
OpenAI o4 Mini (High Effort)
55.0%
#40
Claude Sonnet 4.5
51.0%
#41
Kimi K2
51.0%
#42
GPT OSS 120B
51.0%
#43
Grok 3 Mini
50.0%
#44
Gemini 3.0 Flash
48.0%
#45
Claude 3.7 Sonnet
48.0%
#46
GPT-5.4
47.0%
#47
Gemini 2.5 Flash
46.0%
#48
Llama 4 Maverick 17B
46.0%
#49
Claude 4.0 Sonnet
44.0%
#50
Claude Haiku 4.5
44.0%
#51
GPT-5 Nano (Thinking)
42.0%
#52
DeepSeek V3 (Mar 2025)
41.0%
#53
Qwen 3 Max Preview
40.0%
#54
DeepSeek V3.2
39.0%
#55
OpenAI o3 Mini (High Effort)
39.0%
#56
GPT-5.2
38.0%
#57
GLM-5
37.0%
#58
GPT-5 Mini
36.0%
#59
Claude 4.0 Opus
36.0%
#60
GLM-4.7
36.0%
#61
Mistral Large 3
35.0%
#62
Claude 4.0 Opus (Thinking)
34.0%
#63
Qwen 3
31.0%
#64
MiMo V2 Flash
31.0%
#65
Devstral 2
30.0%
#66
GPT-5.4 Mini
30.0%
#67
DeepSeek V3
29.0%
#68
Gemini 2.0 Flash
28.0%
#69
GPT-5.4 Nano
25.0%
#70
Devstral Small 2
24.0%
#71
Llama 3.1 405B
24.0%
#72
Claude 3.5 Haiku
23.0%
#73
Mistral Medium 3.1
20.0%
#74
GPT-5 Nano
20.0%
#75
Cohere Command A
18.0%
#76
Grok 4.20
18.0%
#77
Llama 3.3 70B
15.0%
#78
Mistral Large 2
5.0%
#79
Magistral Medium 3.1
0.0%
#80
Phi-4
0.0%